Table 5.
Diagnostic accuracy of deep learning-based model predicted diagnosis and experienced radiologist gave a diagnosis
| Tumor group | Sensitivity | Specificity | PPV | NPV | Accuracy | |
|---|---|---|---|---|---|---|
| Brainstem glioma | Predicted by model | 88.24% | 96.23% | 88.24% | 96.24% | 94.29% |
| Diagnosis by radiologist | 100% | 98.11% | 94.44% | 100% | 98.57% | |
| Ependymoma | Predicted by model | 64.29% | 96.43% | 81.82% | 91.53% | 90% |
| Diagnosis by radiologist | 92.86% | 96.43% | 86.67% | 98.17% | 98.18% | |
| Medulloblastoma | Predicted by model | 61.9% | 100% | 100% | 85.96% | 88.57% |
| Diagnosis by radiologist | 95.24% | 100% | 100% | 98% | 98.57% | |
| Pilocytic astrocytoma | Predicted by model | 100% | 78.85% | 62.07% | 100% | 84.29% |
| Diagnosis by radiologist | 88% | 97.62% | 94.12% | 95.35% | 95% | |